Questions on LOD in Models
by Ironbelly Studios · in Artist Corner · 10/14/2008 (9:14 pm) · 7 replies
The first question being if there are any General rules you guys have learnt when setting up your LODS? In my case it would be for an FPS
For example lets say that the highest LOD is 5000 poly's, what do you drop it down to, or at least aim for, for your second highest level? 50%? 75%? and for the third and fourth?
What distances/pixel sizes do you set for each LOD ?
Another question would be how many LOD's do you normally put into a model, what is a good min number to have in there? Can you put too many in?
I was reading that Animations only get played for the top level LOD. Does that mean that once the model is X distance away from the camera and drops down to the second LOD that the animations will stop playing all together? Meaning the model will move through game space but just be totally un-animated whilst doing it.
Thanks in advance and I look forward to hearing what you guys say!
Cheers
For example lets say that the highest LOD is 5000 poly's, what do you drop it down to, or at least aim for, for your second highest level? 50%? 75%? and for the third and fourth?
What distances/pixel sizes do you set for each LOD ?
Another question would be how many LOD's do you normally put into a model, what is a good min number to have in there? Can you put too many in?
I was reading that Animations only get played for the top level LOD. Does that mean that once the model is X distance away from the camera and drops down to the second LOD that the animations will stop playing all together? Meaning the model will move through game space but just be totally un-animated whilst doing it.
Thanks in advance and I look forward to hearing what you guys say!
Cheers
About the author
#2
10/14/2008 (10:10 pm)
So generally 4 levels does it for you? From the research I've been doing you definitely are far from alone in your troubles with getting the LOD to work properly :) 
#3
I'm still playing with the distances on the LODs. Right now I have 0 and 500 for the first and second. The second pops a bit noticeably but its far enough away not to notice too much of the texture distortion.
If you can afford Zbrush it might make sense for your organic models. It more or less has a built in LOD ability. I've just started using the program but it seems powerful. This would let you model at say, 10k tris and then one touch drop down to less. It also handles UV mapping better to avoid headaches LODing can cause there...
With animation you need to rig all the LOD models prior to animating. I usually just overlay all the meshes and get to rigging. It's a bit of a mess but doable with 3 LODs.
It all depends on your game type too. In my case there could be 100 bots on the screen at once. So LOD will become important. I'm not sure how many tris the stock Kork model has in TGE, but I was able to get about 100 on the screen at once (I don't think it has LODs) on my Geforce 6800 and still maintain a good framerate. With that many things running around its more about the netcode and the physics and the AI than the actual GPU load ;0
If you aren't going to be so demanding of the engine you can probably get away with having just 2 LODs, one close and one faraway.
10/14/2008 (11:48 pm)
I usually aim for 5000-7500 tris for my main LOD, then about 3500 for the second and 2000 for the third. My problem is that at some point the model distorts horribly and wouldn't make sense to continue with.I'm still playing with the distances on the LODs. Right now I have 0 and 500 for the first and second. The second pops a bit noticeably but its far enough away not to notice too much of the texture distortion.
If you can afford Zbrush it might make sense for your organic models. It more or less has a built in LOD ability. I've just started using the program but it seems powerful. This would let you model at say, 10k tris and then one touch drop down to less. It also handles UV mapping better to avoid headaches LODing can cause there...
With animation you need to rig all the LOD models prior to animating. I usually just overlay all the meshes and get to rigging. It's a bit of a mess but doable with 3 LODs.
It all depends on your game type too. In my case there could be 100 bots on the screen at once. So LOD will become important. I'm not sure how many tris the stock Kork model has in TGE, but I was able to get about 100 on the screen at once (I don't think it has LODs) on my Geforce 6800 and still maintain a good framerate. With that many things running around its more about the netcode and the physics and the AI than the actual GPU load ;0
If you aren't going to be so demanding of the engine you can probably get away with having just 2 LODs, one close and one faraway.
#4
It's a slow back/forth process, exporting the details>>then viewing inside the engine to get the numbers right, but again, to me, once you get past the first level of detail[0], all the other levels geometry look pretty much the same.
I try to get my highest detail level popping into view just beyond clear visual distance[as a first step], then work back from there. Even without detail levels, as an object recedes into the distance, it visually changes for me. I can see the video drivers crunching away pixels, drawing the object smaller and smaller. I use this 'event horizon' as my benchmark...I get the object at a distance where it clearly comes into view, that's my largest detail level value.
...my 2 cents.
10/15/2008 (5:36 am)
I agree with Jason. I usually put no more than 3/4 detail levels, as once the shapes get far enough away to pop the next level into render, I pretty much loose all visual details after that. Of course, the larger the shape, the more of the detail levels you'll see as the shape recedes into the background....It's a slow back/forth process, exporting the details>>then viewing inside the engine to get the numbers right, but again, to me, once you get past the first level of detail[0], all the other levels geometry look pretty much the same.
I try to get my highest detail level popping into view just beyond clear visual distance[as a first step], then work back from there. Even without detail levels, as an object recedes into the distance, it visually changes for me. I can see the video drivers crunching away pixels, drawing the object smaller and smaller. I use this 'event horizon' as my benchmark...I get the object at a distance where it clearly comes into view, that's my largest detail level value.
...my 2 cents.
#5
LOD_0 = 100% (obviously)
LOD_1 = 75% (loose details like finger definition, nostrils, lips, all the little details)
LOD_2 = 35% (really streamline the shape but keep the form)
LOD_3 = 15% (Mr.Stickfigure! Lose hands and feet totally)
LOD_4 = 8% (straight arms and legs, no elbow or knee joints, just a vague silhouette on the horizon)
My stuff is pretty low poly to begin with, say 2000 tris. And I optimize for proper battlefield ranges of 200-300 metres, which you don't get in most FPS. Most commercial FPS are very close quarters (50-100m).
10/15/2008 (7:23 am)
The way I've been LODing my player/AI models:LOD_0 = 100% (obviously)
LOD_1 = 75% (loose details like finger definition, nostrils, lips, all the little details)
LOD_2 = 35% (really streamline the shape but keep the form)
LOD_3 = 15% (Mr.Stickfigure! Lose hands and feet totally)
LOD_4 = 8% (straight arms and legs, no elbow or knee joints, just a vague silhouette on the horizon)
My stuff is pretty low poly to begin with, say 2000 tris. And I optimize for proper battlefield ranges of 200-300 metres, which you don't get in most FPS. Most commercial FPS are very close quarters (50-100m).
#6
10/15/2008 (8:59 am)
Do you guys find that you prefer to work with distances for your LOD calculations as apposed to pixel size? And if so what distances do you have set for each level? I imagine it will be different for close quarter as apposed to open spaces, our product will be fairly close quarter mostly urban settings.
#7
Best thing to do is put your models out at set distances (20m, 50m 100m, etc) , take a screenshot and then measure the heights in pixels, that way you'll have something to base your LOD changes on.
I'd suggest basing your LODs on the highest resolution you can, then people with lower spec PCs, who tend to use smaller resolutions than 1680x1050, get a performance boost as their LODs kick in earlier. Try to aim for perfection on a high spec PC, but playability on a lower one.
10/15/2008 (1:42 pm)
I think the engine bases it's LOD changes on height in pixels from maximum screen resolution. I don't think you can just base the LOD on distance, not without some hefty engine changes. Best thing to do is put your models out at set distances (20m, 50m 100m, etc) , take a screenshot and then measure the heights in pixels, that way you'll have something to base your LOD changes on.
I'd suggest basing your LODs on the highest resolution you can, then people with lower spec PCs, who tend to use smaller resolutions than 1680x1050, get a performance boost as their LODs kick in earlier. Try to aim for perfection on a high spec PC, but playability on a lower one.
Torque Owner Gary Willhight
Default Studio Name